AfiniTwin: what it is and why any serious AI needs one

An AfiniTwin is a portable cognitive profile package. It’s built by combining six classic psychometric instruments (Big Five IPIP-NEO, AAP, AHP, AVI-25, ATP and a proprietary conversational module) with a standard injection protocol. The output is a file you paste as a system prompt into any AI interface and it conditions the conversation from the first message on.

What you get is not a simplified version of you. It’s the same representation Afini.ai uses internally to make its own assistant treat you as a concrete interlocutor — with your dissent thresholds, your ambiguity tolerance, your humour, your time horizon — only now downloadable and reusable outside the platform.

The difference between having an AfiniTwin and not is the difference between an assistant that knows you and one that probes you with soft questions every time you start a thread.

Commercial LLMs are trained to be agreeable by default. It’s learned behaviour — not malice, not bug. But it produces a painful bias for anyone who uses them to decide things: they validate more than they push back, soften more than sharpen, and when they should warn you of a risk they offer three balanced alternatives for you to pick.

That works for 95% of casual uses. It fails right where you most need the AI: in a hard decision, in an honest critique of your own writing, in analysing a conflict you’re emotionally inside.

The AfiniTwin corrects that asymmetry by injecting two things the model doesn’t know: who you are (psychometric profile) and how you want to be treated (modes, meta preferences, explicit contraindications like anti-sycophancy or emotional suppression).

«Your AI isn’t a flatterer because it likes you. It’s trained to be friendly to anyone. Your AfiniTwin recalibrates it to be friendly to you in particular — which sometimes means saying something you won’t enjoy.»

The package you download is not a PDF or an ornamental document. It’s a structured set of artefacts designed to be injected as operational context in any serious LLM.

• Cognitive profile as structured JSON — the six psychometric instruments compressed into a schema any agent can parse.
• Narrative system prompt in six languages — the text version you paste straight into Claude, ChatGPT, Gemini or Mistral.
• Eight mode variants (Interlocutor, Voice, Critic, Analyst, Mirror, Devil’s Advocate, Translator, Projective) — each tuned for a specific use.
• Platform-specific templates — formats for Claude Projects, ChatGPT Custom GPTs, Gemini Gems, API system instructions.
• Interpretation protocol — a self-contained document that teaches the LLM how to read your profile without distortion (how to manage sycophancy bias, the trait/state distinction, the contraindications).
• Permanent snapshot — a permalink to your profile state at purchase time, in case you lose it or it gets updated.

What an AfiniTwin is not matters as much as what it is. Three things you won’t find in the package:

• It’s not an alternative personality you paste in to make the AI pretend to be you. It’s context on how to treat you, not a costume.
• It’s not a clinical diagnosis. The instruments we use are psychometric traits, not pathologies or DSM labels. We won’t tell you you have anxiety or suggest treatments.
• It’s not a prediction of the future. The profile describes stable, declared patterns; the decision is still yours. Your AfiniTwin doesn’t decide for you — it calibrates the AI so that when it decides with you, it does so on your reality, not the statistical average.

If you’re happy with an AI that validates almost everything and you only use it for mechanical tasks (summarising documents, translating, writing low-stakes emails), the AfiniTwin won’t change your life. The price isn’t justified.

Where it is justified is when the AI is part of your thinking: for writing things you’ll sign, deciding things you’ll live with, analysing conflicts you’re inside, preparing hard conversations, reviewing your own work with a critical peer who doesn’t sugar-coat.

In those cases, the difference between having an AfiniTwin and not is the difference between an assistant and an instrument.